Randy Bias, CTO and co-founder of cloud consulting firm Cloudscaling, joins us on this week's episode of Cloud Cover TV to share his thoughts on "enterprise clouds." Spoiler alert: he thinks they're doomed to failure.
We also discuss Cisco's purchase of NewScale and the new "dedicated instances" from Amazon Web Services.
For the rest of the episodes, check out the Cloud Cover TV home page.
Read the full transcript from this video below:
Randy Bias dumps on enterprise clouds
Jo Maitland: Welcome to Cloud Cover TV, our weekly show on
all the juiciest news in the cloud computing market. I am Jo Maitland,
in San Francisco. This week our special guest on the show is Randy Bias,
founder and CTO of Cloud Scaling. Cloud Scaling is a boutique consulting
company here in the city that provides service providers and enterprises
with the help to build infrastructures as a service, on top of their existing IT.
The company's list of customers include KT Telecom, the largest carrier in
Korea; Cloud Central, a service provider in Australia; and Engine Yard, a
platform as a service company here in the US, among a few others.
The list is small, but Cloud Scaling’s influence in the industry is growing,
largely due to Randy. He gave a presentation at the Cloud Connect
Conference recently, and has some strong views on enterprise clouds,
meaning public or private clouds based on legacy technology from the
traditional IT vendors, and why these in his belief, will fail. We asked him
on the show to explain what he means by enterprise cloud and why he
thinks these are doomed.
Randy, tell us about this idea that you have been talking about, in terms
of enterprise clouds versus the public cloud. What exactly do you mean
by enterprise cloud, and you talk about that as if it is doomed.
Randy Bias: The shortened answer is basically, that if you look at what
I call an enterprise cloud, it is a way of building a cloud that resembles
traditional enterprise data centers that we have today, as opposed to a new
way of building cloud, what I call cloud computing, which is personified in the
way that Amazon has built their cloud. That is really how I would differentiate
Jo Maitland: From a technology perspective, what exactly are we talking
about? Commodity hardware versus expensive hardware? Open source
software versus legacy, expensive software?
Randy Bias: There seems to be this historical drive to take business
infrastructure, whether it is telecommunications, railroads, air transportation,
shipping logistics, financial services, banking services, and to commoditize
them over time, and that pressure to commoditize business infrastructure is
what I am referring to. One of the ways we see Amazon being successful
at commoditizing is taking all of the dollars, the costs out of the stack. You
see this with Amazon and Google, they reduce their data center costs for power,
for cooling, for electricity, for the cost of the servers inside of it, and also for the
software inside of it. So, yes, commodity hardware, open source software, and
a focus on continuing to drive all the costs related to convenient storage and
network closer to zero.
Jo Maitland: One of the things that we hear from the IT side is that there
is actually a lot of value in things like the VCE System, which is the product
from VMware Cisco EMC. The integration that they have done of those big
pieces of infrastructure and the lessening the chance of single points of
failure in these silos of IT, they have created value there, and that is very
valuable, number one; number two, they are trusted sources of infrastructure.
Randy Bias: Sure.
Jo Maitland: I wonder how you would help the organizations
think about that.
Randy Bias: There is a lot to crack into there. There is a lot of ways
that we could approach that. The first thing I would say is that if we
look at any kind of disruption that happens, if we look at any new way
of doing any kind of technology, then it is never the incumbents that are
the leaders. Look at Cisco, Cisco created and basically drove the adoption
of TCPIP, and in large part was one of the drivers of success of the rapid
deployment of the internet. Before that, networking existed, XDOT 25, all
kinds of other technologies It would not have worked if there was not a startup
like Cisco to help create that disruption. To look at the leaders or today's
incumbents and assume that they are going to help us get there, history does
not tell us that, that is how it happens. That is why startups exist, and that is why
they are successful against very large incumbents, traditionally.
The second part of this is, there is an over-focus in redundancy, and
you see this, this is a very enterprise computing mindset. When we think
about cloud computing, we think about it as a new way of doing IT. We
think about it as a type of IT that displaces the current dominant paradigm
of enterprise computing, in the same way that enterprise computing
displaced mainframe computing. One of the dominant design patterns in
enterprise computing is these pairs of HA systems. You see this throughout
all traditional enterprise data centers, there have pairs a load balancers,
firewalls, switches, routers, servers, et cetera, et cetera. There is a couple
problems with that. The first is that it relies on a strategy that is about 100%
risk - mitigation, it attempts to remove all failure point in a system, but we know
that at scale, everything fails. What Amazon or Google have done completely
differently is they have built systems that assume failure happens, designed for
failure model, and actually design around that. Somebody like Google does not
bother to make sure that all of their servers are up all the time, in fact 25% to 30%
of their servers are down and have hardware problems at any given time. That
design pattern does not exist inside of the enterprise data center, where people
try to have HA pairs, and remove all failure points.
The second part of that HA, is that you are basically paying twice for the
same capacity. For a certain amount of money, an enterprise would buy
two very expensive HA boxes. Google will buy ten extremely inexpensive
commodity boxes, and design their software in such a way that a third to half
of those can be down at any given time, and they have no performance impact,
no service impact, and their costs for ten boxes that are probably close to 2 to
4x more performance than the two HA boxes, is about the same as the two
Jo Maitland: Do you think the Cisco/Amazon data center requires more or
less people? I notice the Amazon website services is hiring people by the
Randy Bias: Boatloads
Jo Maitland: By the boatloads.
Randy Bias: It requires much less people; that is the short answer. One
of the other design patterns, or at least emergent behavior in enterprise
computing is that the data centers are extremely heterogeneous, so there is
all kinds of systems, all kinds of different storage systems, and networking
servers and appliances of various kinds. Inside Amazon or Google, things are
much more homogenous, it is not completely homogenous, but it is significantly
more homogenous, and that allows for much greater operational efficiency. You
do not have to train people on five or six different storage systems, you just have
one. In Google's case, they simply use a software tier called Google FS, which is
their single, global distributed file storage system.
Jo Maitland: Let us say I am a more forward-looking IT organization.
I decided, ‘Right. That is interesting, it is disruptive. Let us start up a little
group inside of our IT division to build this kind of infrastructure.’ Am I
looking at sourcing white boxes from Taiwan, should they be looking at
open source? What are all these components? That is the problem, Amazon
Web Source and Google are not very open about what their infrastructure is
built on. Where do we go for this stuff?
Randy Bias: I do not really think there is any source at the moment. I
think that will change. I know of projects that are in flight that are going
to change that, but right now you are sort of stuck. There is no way to
go by Amazon or Google, but perhaps more importantly is that I notice
that even if you buy into this, there are structural and cultural problems.
I have sort of been dinged in the past about talking about those over the
technology issues, but what we have seen is that they are the bigger problem.
Technology actually, is not the hardest problem. If you were to go look at
Amazon.com, or Amazon Web Services, the way that they do operations and
IT, think about things, deliver services, structure their teams, are completely
and utterly different from an enterprise IT shop. I mean, just completely alien.
One of the problems that I see, as an example, is the typical decision making
process for making purchasing decisions inside of an enterprise is what I call
the CYA process, cover your a**. The challenge is when you look at a service
provider, anybody who is making money off the IT services that they build and
provide, they make an ROI decision, so you are talking about an ROI decision
versus a CYA decision. CYA decisions drive decisions like, 'Well, nobody ever
got fired for buying IBM. Nobody ever got fired for getting Cisco and VC.' Those
decisions will not result in success; they will result in failure, whereas ROI-based
decisions will result in success. If you can get the executive-level sponsorship
for you to build a cloud like Amazon Web Services and Google, and get them
to buy in that you are actually going to deliver ROI, and that you are going to
be very focused on something that can be competitive with Amazon, but
deployed internally, I think that is the willpower that would allow you to succeed.
I have not seen one enterprise do that yet today. I see teams that are running
off and trying to figure out how to do it without an executive-level empowerment,
but without that, I think that it is going to be a very difficult path for them.
Jo Maitland: In other news this week, Cisco announced it will buy New Scale,
a startup here in the valley, which sells software that lets IT organizations and
service providers build a service catalog on top of their infrastructure, and offer
a portal that lets consumers of IT services request their own services based on
this catalog. The self-service component is actually one of the big differentiators
between cloud and virtualization. For all of you out there who think you have
cloud because you have virtualization, afraid not. Consumers of IT services must
be able to trigger deployment of services on their own. This is what New Scale
does, and it has about 100 customers using its software today, including Allied Irish
Banks, Marriott, and a couple of insurance companies, Mutual of Omaha being
one of them.
At the orchestration layer, Cisco is also partnered with BMC Software
back in December, to build something called the Integrated Cloud
Delivery Platform. This jams BMC's orchestration product, called Cloud
Lifecycle Management, on top of Cisco's integrated server and networking
infrastructure. We have not had anyone building infrastructure as a service
yet, using these products, though. Cisco is also a partner in the BCE
Company, which sells a pre-integrated stack of hardware that includes
EMC storage, Cisco servers and networking, and VMware virtualization.
To Randy's point earlier, this is the expensive approach to building cloud
infrastructure. It is not clear yet from Cisco what the company's vision really
is for cloud computing, except to cast a wide net with partnerships and
acquisitions, to see what sticks.
On to public cloud news briefly. AWS, the giant online book retailer unveiled
yet another way to buy its EC2 instances called Dedicated Instances, that is
the new purchasing model. These dedicated instances let users buy isolated
hardware, physical hardware, just for their use, locked in through a virtual
private network. Typically, the cloud model is multi-tendency, in this instance
you are buying your own iron. It is somewhat of a backpedal by Amazon Web
Services, as the company curtails its service offerings to meet the security and
availability needs of enterprise IT departments. These guys are much less
interested in elasticity and super low prices, and more interested in security
and reliability. You will pay for it, though. These dedicated instances cost an
extra $10 per hour, per region, on top of $0.105 per hour for the small instances,
running Unix or Linux, and $0.145 per hour for Windows. That works out as a
$0.02 premium on the instance over the usual price, plus the $10 per hour. It is
all about the costs here with AWS, and supposedly making things cheaper for
you. With these dedicated instances, the question is, can IT departments buy,
run, and support service for themselves, cheaper? We will see.
This has been Cloud Cover TV. Thank you for watching. Tune in next week
for more insider news on the cloud computing market.